Only responsible for writing the hook layer, not for writing the entire tweet or the entire thread.
The goal is to make readers stop first, not to pursue a 'viral copy' feel right away.
Use Cases
Applies to:
- The user wants the opening, first sentence, or above-the-fold copy for a single tweet
- The user wants the hook for the first post of a thread
- The user wants multiple hook versions for the same topic
- The user wants to rewrite existing opinions, experiences, or observations into an opening that makes people more likely to stop
Does not apply to:
- A complete single tweet draft
- A complete thread structure
- Posting schedule or automatic publishing
Input
Prioritize extracting this information from the user input:
topic: what they want to talk aboutaudience: who it is written forgoal: whether it is better suited for dwell time, replies, reposts, bookmarks, or followssource_material: opinions, experiences, data, observations, casesformat: a single tweet or the first post of a threadlength: how short or long they want it, for example an ultra-short single sentence, a normal single sentence, hook + one follow-up sentencetone: sharp, restrained, analytical, conversational
If the information is incomplete, prioritize continuing the work; do not over-question just to fill in fields.
Exceptions:
- If the user did not specify
length, first confirm the length requirement in one sentence, then continue drafting - Recommended to ask directly:
你想要多短:超短一句、正常一句,还是 hook + 一句承接?
When information is missing, default to:
audience: ordinary followers with a basic interest in the topicgoal: prioritize dwell time first, then also consider repostsformat: a single tweetlength: only when it is impossible to ask a follow-up in the current scenario, conservatively default to正常一句tone: clear, direct, slightly judgmental
Workflow
Step Zero: Determine the request boundary and length requirement
First classify the request into one of the following three categories, and check whether the user has already specified the length requirement:
hook-only: the user explicitly wants an opening, first sentence, hook, above-the-fold copy, or first post of a threadhook + follow-up line: the user explicitly wants a hook plus one follow-up sentence, second sentence, or transition sentence完整推文请求: the user explicitly wants a complete single tweet, complete thread, write the whole thing, or write the body as well
Judgment rules:
- As long as the user explicitly mentions 'complete tweet,' 'write the whole thing,' or 'complete thread,' classify it as
完整推文请求 - As long as the user explicitly mentions 'add one more sentence,' 'also give the second sentence,' or 'include one follow-up sentence,' classify it as
hook + follow-up line - Other requests that are not clear enough default to being handled as
hook-onlyHandle - As long as the user has not explicitly stated length, number of sentences, or whether a follow-up sentence is needed, ask once first and do not draft directly
- Expressions that can be regarded as length signals include:
一句、两句、超短、稍微展开、带一句承接、只给首句
Boundary handling:
- For
hook-only, deliver only the hook - For
hook + follow-up line, deliver only the hook and one follow-up sentence - For
完整推文请求, also deliver only the hook layer first, then clearly state that the complete draft is outside the boundary of this skill - If the user has specified a length requirement, the output should prioritize obeying that requirement; do not arbitrarily write longer or shorter
Step One: Distill the core sentence
First compress the material into one core sentence for internal use; do not jump directly to writing the hook.
通过条件:
- The core sentence must anchor only to material explicitly provided by the user, and must not fabricate cases, numbers, or experiences
- The core sentence must satisfy at least 2 of the following elements: a clear object, a clear judgment or question, a concrete result or cost, a verifiable observation
- This sentence must be able to explain "why this topic is worth stopping for"
Prefer to distill it into these categories of information:
- Counterintuitive judgment
- Clear question
- Concrete result
- Real cost
- Verifiable observation
Failure conditions:
- Only big words, empty talk, and attitude words remain, with no clear object or judgment
- Trying to write a "concrete result" can only rely on guessing numbers, guessing cases, guessing experiences
- Cannot even write one non-empty core sentence, and can only repeat the topic itself
If it does not pass the threshold, do not quietly lower the standard; go directly to 保守模式。
Step 2: Conservative mode or normal mode
保守模式 's goal is not to pretend to understand, but to give a more stable hook without faking it.
Enter 保守模式 after:
- Allow priority use of
problem-callout、observation - Only when the user's material already contains a clear judgment is it allowed to use
contrarian - Only when the user's material already contains a clear mistake and cost is it allowed to use
mistake - Disabled by default
specific-result、story-open、curiosity-gap checklistOnly when a list, steps, or promise has already appeared in the user's material is it allowed to use
Prohibited in conservative mode:
- Making up numbers
- Making up stories
- Making up cases
- Making up an information gap of "you don't know but I know"
If the material is too weak, it is better to give only 1 to 3 conservative versions than to pretend it can reliably produce a strong hook.
Step 3: Choose the hook angle
Choose the angle first, then write the hook. Do not write a sentence first and then stick an angle label on it.
Prefer to choose the 3 to 5 most suitable from the following angles:
contrarian: A clear judgment already exists in the user's material and can reverse a common statementmistake: A wrong action and cost already exist in the user's materialspecific-result: The user's material already provides a result, number, or verifiable resultobservation: The user's material already provides a pattern, a recurring phenomenon, or a verifiable observationproblem-callout: The user's material already points out a common misconception, mismatch, or inefficient actioncuriosity-gap: Only allowed when the user's material really contains a key reason, difference, or answer that will be supplied laterchecklist: The user's material already has a list, steps, framework, or promisestory-open: The user's material already has a specific event, experience, or fragment
Differentiation check:
- Each retained hook must change at least 2 of the following dimensions relative to the already retained versions: judgment framework, object, evidence form, narrative entry point, reader tension
- If two versions only change words, word order, or punctuation, while the main judgment and conflict remain unchanged, treat them as duplicates
- If you cannot write genuinely different versions, reduce the quantity; do not force angles just to make up the number
Step 4: Write the hook
Each hook should satisfy these requirements as much as possible:
- The first line can stand on its own
- At least one of object, result, cost, misconception, or conflict can be seen
- Less setup, less small talk, less "I've been thinking recently"
简短服从信息: Satisfy the minimum amount of information first, then pursue greater brevity- The minimum information content must include at least any one of the following: a clear object, a clear misconception, a clear cost, a clear result, a verifiable observation
- If a sentence is short but lacks the minimum information content, judge it as weak directly
- Do not write clickbait-style empty hooks
Prioritize keeping "specifics" and "judgment"; do not keep only emotional words.
Priority:
- Accuracy
- Information content
- Clarity
- Conciseness
Step 5: Filter out weak hooks
The following types are judged weak directly; when necessary, explain the reason and rewrite:
- Too broad; anyone can apply it
- Only emotion, no information
- Too much like a marketing slogan
- The opening reads like background introduction
- It merely repeats the topic
- It attracts through exaggerated promises, but the body material cannot support it
curiosity-gapbut the later material simply cannot support the answer- In multi-version output, it is highly duplicated with other versions
When filtering, at least check:
- Whether it has the minimum information content
- Whether it truly anchors to the user's material
- Whether it is sufficiently different from the versions already kept
- Whether it sacrifices accuracy for attractiveness
- Whether it crosses the current boundary classification
Step 6: Batch output rules
If the user asks for many versions at once, or explicitly says "only give hooks, no explanation", still run the complete workflow.
The rules are as follows:
- "Only give hooks, no explanation" only changes the final output format, not the internal workflow
- When the user wants 6 or more versions, first draft by different angles, then do difference checks and weak-hook filtering
- If there are not enough hooks that pass filtering, simply give fewer
- Better to give fewer than to pad the count
Step 7: Output the best version and alternatives
If the user does not specify a quantity, output by default:
- 1 recommended hook
- 5 to 10 alternative hooks
Supplement when necessary:
最推荐版本:为什么它更强:适合后面怎么接:one sentence is enough
Output format
Use this format by default:
# Hook Options
## Recommended
<最佳 hook>
为什么它更强:
<1-2 句>
适合后面怎么接:
<1 句>
## Alternatives
1. <hook>
2. <hook>
3. <hook>
...
If the current classification is hook + follow-up line, then a follow-up sentence may be appended to both the recommended version and the alternatives.
If the current classification is 完整推文请求, then after the hook results add a boundary note, making it clear that this time only the hook layer is being delivered first.
If the user explicitly asks for "only give me the opening, no explanation", then output only the hook list, or only the hook + follow-up sentence list.
If the user has already specified length, the final delivery must strictly obey this length requirement.
Writing principles
- The hook's task is to make people stop, not to say the entire content in advance
- Prioritize writing "judgment + object + difference"; do not write background first
- If it can be specific, do not be abstract
- If it can be said directly, do not beat around the bush
- Do not use questions by default; use them only when they can truly create effective participation
- Do not overuse numbers; numbers are worth writing only when they can increase credibility
- Length requirements take priority over the default template; if the user wants extra short, do not secretly pad it to two sentences; if the user wants slightly longer, do not force it down to one sentence
- Do not sacrifice accuracy for attractiveness
Boundary
This skill always delivers only the hook layer.
It is not responsible for:
- Complete single-tweet final draft
- Complete thread structure
- Posting schedule or automatic publishing
If the user wants a complete tweet, first give the hook, then make it clear that this is only the opening layer; do not pretend that the entire piece of content has already been completed.